The diversity of land surfaceleads to an increase of the uncertainty in surface reflectance calculation, whichreduces the-accuracy of land-atmospheric decoupling, and then of the aerosol retrieval. The multi-spectral, multi-angle and polarized information, which can be provided by the Atmosphere Multi-angle Polarization Radiometer (AMPR), is helpful to solve the problem. The behavior of aerosol scattering amongdifferent wavelengths is smooth. At the same time the land surface polarized reflectance does not show any dependence in the measurement bands. Based on these conclusions, we developed a method to separate land surface and atmosphericradiation by successive atmospheric correction. Further on, the aerosol retrieval algorithm was build. In the algorithm, the aerosol optical thickness was retrieved at 665 and 865 nm, while thesurface polarized reflectance was reobtained at 1 640 nm by atmospheric correction. The atmospheric correction (1 640 nm) and aerosol retrieval (665 and 865 nm) form a cycle. Then, an iteration method was employed and it approaches to the real radiant values of atmosphere and land surface step by step. At last, we retrieved aerosol optical depth from a look-up table which was builtby 6 fine and 6 coarseaerosol modes. The AMPR have performedat least 5 flights mission over Beijing, Tianjin and Tangshan region. Data of theseobservations were used to verify the algorithm and it was found that the AOD retrieved from AMPR and CE318 have good coherence. When the AOD is lower than 0.5, the average deviation is below 0.03.